Energy conservation and emission reduction in steelmaking have received significant attention owing to the high amount of fossil energy consumption and emissions. Many methods have been adopted for saving energy, among which coke dry quenching (CDQ) is a cost-effective option. In this work, a CDQ process in a steel plant in China is studied. Here, an economic efficiency index is adopted to handle the trade-off between the steam productivity and the coke burning loss. The operation data analysis indicates that the supplementary air flow rate in the CDQ operation does not follow the variation in the discharge rate of incandescent coke adequately, and this results in an increase in the concentration of combustible gas in the exhaust gas and a decrease in economic efficiency. The correlation analysis results show that it is necessary to introduce several derived variables into the data-driven model of this process because these derived variables are more useful than a few original variables for the prediction purposes. Based on these analyses, a soft sensor is proposed by integrating a nonnegative garrote variable selection algorithm with an autoregressive integrated moving average model, which provides a good solution for the real-time prediction of the economic efficiency of the CDQprocess. Using this soft sensor, model-based optimization can be conducted, the performance of which is verified with a virtual implementation on the historical operation data and experiments performed in a real CDQ system. The results indicate that there is considerable room for improving the economic efficiency of this process. (C) 2019 Elsevier Ltd. All rights reserved.